RNA-Seq Technology and Its Application in Fish Transcriptomics

Abstract

High-throughput sequencing technologies, also known as next-generation sequencing (NGS) technologies, have revolutionized the way that genomic research is advancing. In addition to the static genome, these state-of-art technologies have been recently exploited to analyze the dynamic transcriptome, and the resulting technology is termed RNA sequencing (RNA-seq). RNA-seq is free from many limitations of other transcriptomic approaches, such as microarray and tag-based sequencing method. Although RNA-seq has only been available for a short time, studies using this method have completely changed our perspective of the breadth and depth of eukaryotic transcriptomes. In terms of the transcriptomics of teleost fishes, both model and non-model species have benefited from the RNA-seq approach and have undergone tremendous advances in the past several years. RNA-seq has helped not only in mapping and annotating fish transcriptome but also in our understanding of many biological processes in fish, such as development, adaptive evolution, host immune response, and stress response. In this review, we first provide an overview of each step of RNA-seq from library construction to the bioinformatic analysis of the data. We then summarize and discuss the recent biological insights obtained from the RNA-seq studies in a variety of fish species.

Introduction

Transcriptomics refers to the study of the complete set of transcripts in a specific cell, tissue, or organism for a given developmental stage or physiological condition (Wang et al., 2009). This complete set of transcripts is known as a transcriptome, including protein-coding messenger RNA (mRNA) and non-coding RNA [ncRNA: ribosomal RNA (rRNA), transfer RNA (tRNA), and other ncRNAs] (Lindberg and Lundeberg, 2010; Okazaki et al., 2002). Unlike the relatively stable genome, the transcriptome varies with developmental stage, physiological condition, and external environment. Transcriptome analysis is a powerful tool for dissecting the relationship between genotype and phenotype, leading to a better understanding of the underlying pathways and mechanisms controlling cell fate, development, and disease progression. The aims of transcriptomics are not limited to the quantification of change in expression level for each gene among different transcriptome samples but include also the mapping and annotation of the transcriptome and the determination of the functional structure of each gene in the genome (Costa et al., 2010; Ruan et al., 2004; Wang et al., 2009).

The complexity of transcriptome determines the reliance on high-throughput tools for transcriptome studies. Over the years, several technologies, either hybridization- or sequence-based, have been developed to survey transcriptomes in a high-throughput manner. Hybridization-based technologies usually rely on incubation of fluorescently-labeled cDNA with probes fixed onto a solid surface (microarray). Since its first application in 1995 (Schena et al., 1995), microarray has been widely used in transcriptomics. Tilling microarray, an updated microarray with probes representing the genome at a high density, has also been generated and can interrogate transcriptomes at a relatively high resolution and can even discover novel transcripts (Cheng et al., 2005; David et al., 2006; Kampa et al., 2004). The use of microarray in fish biological studies has been more than one decade and well reviewed elsewhere (Douglas, 2006; Hook, 2010; Nielsen and Pavey, 2010; Zhang et al., 2009). Microarray technology, however, suffers from some intrinsic limitations (Bradford et al., 2010; Fu et al., 2009; Marioni et al., 2008; ’t Hoen et al., 2008), including dependence on the existing knowledge of genomic sequence, signal saturation for certain transcripts with a high abundance, and high background noise due to nonspecific hybridization. Aside from these intrinsic disadvantages, specific problems exist for fish transcriptomic studies when using microarray: only a few commercial microarray platforms are available, and they are specifically designed for model species such as zebrafish (Danio rerio) (Douglas, 2006); the construction of customized microarray or tiling array relies on the preexisting knowledge about EST sequences or genomic sequences, which is usually not available for many fish species (Douglas, 2006; Zhang et al., 2009).

In contrast to microarray, sequence-based technologies directly determine the cDNA sequence. These approaches rely on the generation of featured EST tags that correspond to the fragments of those transcripts in a sample and their subsequent concatenation prior to cloning and sequencing (Fullwood et al., 2009). Different tag-based sequencing methods, such as serial analysis of gene expression (SAGE) (Matsumura et al., 2005; Velculescu et al., 1995), polony multiplex analysis of gene expression (PMAGE) (Kim et al., 2007), and massively parallel signature sequencing (MPSS) (Brenner et al., 2000), have already been developed. SAGE has so far been used in three transcriptomic studies in zebrafish regarding oogenesis (Knoll-Gellida et al., 2006), sexual dimorphism (Zheng et al., 2013), and stress response to toxicant (Cambier et al., 2009). Most of these tag-based sequencing methods, however, are based on expensive Sanger sequencing; these methods are labor-consuming and time-consuming for the cloning step (Morozova et al., 2009); it is often impossible to precisely annotate the tags unless the whole genome information is available (Costa et al., 2010); only a portion of transcripts are analyzed (Harbers and Carninci, 2005); and differential isoform and allelic expressions are usually indistinguishable (Wang et al., 2009).

Recently, the advent of low-cost (NGS) technologies, paralleling the sequencing process, has led to the generation of a new method for both mapping and quantifying transcriptome, known as RNA-seq. RNA-seq is free from almost all the limitations of the methods mentioned above (Table 1). RNA-seq has only been available for several years; however, this method is already revolutionizing the field of transcriptomics, improving our understanding of genome expression and regulation. Importantly, RNA-seq, combined with other state-of-the-art omics technologies, has been applied to analyze the detailed integrative personal omics profile for evaluating disease risk and monitoring disease progression for personalized treatment (Chen et al., 2012; Roychowdhury et al., 2011). Besides, RNA-seq has already been applied to a substantial amount of fish biology studies (Table 2).

Table 1.

Comparison of RNA-Seq with Other Methods for Surveying Transcriptome^*

	Tilling microarray	Tag-based sequencing	RNA-Seq
Principle	Hybridization	Sanger sequencing	Parallelized high-throughput sequencing
Resolution of data	Several to 100 bp	Single base	Single base
Sensitivity	Low	Moderate	High
Throughput	High	Low	High
Turnaround time	Long	Long	Short
Required amount of RNA samples	High	High	Low
Cost per sample (excluding equipments)	High	High	Relatively low
Reliance on existing genomic sequence	Yes	No	No
Linear dynamic range of expression levels	<2 orders of magnitude	Not practical	Limited only by sequencing depth
Discovery of unknown transcribed regions	Limited	Yes	Yes
Detection of differences in isoformic and allelic expressions	Limited	Limited	Yes
Detection of mutations	Limited	Yes	Yes
Determination of splicing sites	Limited	Yes	Yes
Identification of UTRs	Limited	Limited	Yes

^*The information in this table was summarized from Wilhelm et al., 2009 and Wang et al., 2009.

Table 2.

Representatives of Publications that Have Had Fish Transcriptomes Studied by RNA-Seq

Major application	Fish species	Sequecing platform	Reference
Transcriptome mapping and genome annotation	Poecilia reticulate	Roche 454	Fraser et al., 2011
	Danio rerio	Illumina	Collins et al., 2012
	Ictalurus punctatus	Illumina	Liu et al., 2012b
	Nothobranchius furzeri	Illumina	Petzold et al., 2013
Novel transcript discovery	Labeo rohita, Hamilton	Illumina	Robinson et al., 2012
	Danio rerio	Illumina	Pauli et al., 2012
	Oncorhynchus mykiss	Illumina	Palstra et al., 2013
	Salmo salar	Illumina	Kure et al., 2013
Detection of RNA splicing	Danio rerio	AB SOLiD	Aanes et al., 2011;
SNP discovery	Ictalurus furcatus	Illumina	Liu et al., 2011
	Cyprinus carpio	Illumina	Xu et al., 2012
	Oncorhynchus mykiss	Illumina	Salem et al., 2012
Quantification of gene expression	Danio rerio	AB SOLiD	Vesterlund et al., 2011
	Lates calcarifer	Roche 454	Xia et al., 2013
	Lateolabrax japonicus	Illumina	Xiang et al., 2010
	Fundulus heteroclitus	Roche 454	Oleksiak et al., 2011
	Salmo trutta	Illumina	Uren Webster et al., 2013
	Fundulus grandis	Illumina	Garcia et al., 2012
	Perca flavescens	Roche 454	Pierron et al., 2011
	Melanotaenia duboulayi	Illumina	Smith et al., 2013
	Coregonus clupeaformis spp., Salmonidae	Roche 454	Jeukens et al., 2010
	Gasterosteus aculeatus	Illumina	Greenwood et al., 2012
	Astyanax mexicanus	Roche 454	Gross et al., 2013

In this review, we present an overview of RNA-seq method from data generation to bioinformatic analysis, discuss the challenges for RNA-seq, and then review the biological insights already gained from RNA-seq for a variety of fish species. We apologize to all those investigators whose articles were not cited due to space constraints.

Overview of RNA-seq

RNA-seq, also known as whole transcriptome shotgun sequencing (WTSS), employs the NGS technologies to sequence cDNA directly from a RNA sample of interest (Morozova et al., 2009; Wilhelm et al., 2008). Transcriptome analysis by RNA-seq is a three-step method, including library construction, sequencing on a specific NGS platform, and bioinformatic analysis (Fig. 1). In the remainder of this section, we will explain each step of a typical RNA-seq experiment and discuss the accompanying challenges and the possible solutions.

FIG. 1.

Illustration of RNA-Seq data analysis pipeline.

Library construction

The library preparation is a key step for RNA-seq as it determines to a large extent how accurately the final sequencing data reflects the original transcriptome. The first procedure in this step is to collect appropriate samples, usually tissues, to be analyzed. One key concern here is the invariably heterogeneous nature of tissues. This is because tissues typically contain tens or hundreds of unique cell types, and thus, transcriptomic analysis of a tissue confounds the real transcriptomic profiles of its constituent cell types (Islam et al., 2011; Shapiro et al., 2013). One solution to this problem is to analyze single cells rather than cell populations. In fact, various single-cell isolation methods have been developed and already coupled with RNA-seq technology in practice (Shapiro et al., 2013). This single-cell RNA-seq lets many previously impossible applications in both basic and clinical research become possible, such as characterization of the initial differentiation events in embryogenesis (Tang et al., 2009; 2010), the investigation of tumor heterogeneity (Dalerba et al., 2011), transcriptomic analysis of rare, transiently existing adult stem cells (Lister et al., 2011), and others.

Following sample collection, total RNA is usually prepared via organic extraction and/or absorption onto silica-membranes of spin columns. Next, RNA is converted to a library of cDNA fragments. Although total RNA can be directly used, total RNA has to be fractionated in most cases. This is because the rRNA, making up more than 80% of total cellular RNA (Lindberg and Lundeberg, 2010), is usually not the research focus, and its presence greatly reduces the useful transcript coverage in the following sequencing step. Total RNA samples are therefore processed either by direct selection of poly(A) RNA or by selective removal of rRNA (ribo-depletion) (Costa et al., 2010). Oligo(dT)-based mRNA purification procedure, widely used in eukaryotes, takes advantage of the presence of a poly(A) tail at the 3′ of eukaryotic mRNA. A large fraction of non-ribosomal RNAs (both coding and noncoding) in eukaryotes, however, lacks the poly(A) tail and is therefore missed (Jacquier, 2009). When compared to the poly(A) RNA selection, the ribo-depletion method is preferred since it enriches all nonribosomal RNA species, including nonpoly(A) mRNA, preprocessed RNA, tRNA, and other ncRNAs with known or unknown functions (Lindberg and Lundeberg, 2010). Although many different rRNA depletion methods have been developed, the two most popular ones are (He et al., 2010; Wilhelm and Landry, 2009): (1) hybridization capture of rRNA by the biotin-labeled anti-rRNA probes, followed by removal with streptavidin-coated magnetic beads; and (2) selective degradation of rRNA by a 5′-3′ exonuclease that specifically recognizes rRNA with a 5′ phosphate.

A double-stranded cDNA library is then prepared via either RNA fragmentation (RNA hydrolysis or nebulization) prior to the reverse transcription or reverse transcription first followed by cDNA fragmentation (DNase I treatment or sonication) (Roberts et al., 2011; Wang et al., 2009). The purpose of fragmentation is to reach the desired length for NGS technologies (Metzker, 2009). Each of these two fragmentation methods causes a different bias in final results. Particularly, cDNA fragmentation usually generates an under-representation of the 5′ of the transcripts in the data, while RNA fragmentation allows a good representation of the transcript body but causes depleted transcript ends (Mortazavi et al., 2008; Nagalakshmi et al., 2008). Additionally, the process of reverse transcription can also complicate final RNA-seq data due to the tendency of reverse transcriptase to generate spurious second-strand cDNA and the artificial chimeric transcripts introduced by template switching (Ozsolak and Milos, 2011b).

After the generation of fragmented cDNA, sequencing adapters are ligated to both ends of the fragments. During this process, information about the orientation of transcripts is completely lost. Fortunately, strand-directionality information can be maintained by converting cytidine into uridine with sodium bisulfate (He et al., 2008); the resulting C–T transition position then labels the coding strand of each transcript. Other approaches that maintain the strand specificity are involved with how the adaptors are ligated to the cDNA fragments, and these methods are well reviewed elsewhere (Costa et al., 2010; Marguerat and Bähler, 2010). Like the process of fragmentation, adaptor ligation may also introduce coverage non-uniformity since the fidelities of DNA ligases cannot be guaranteed (Faulhammer et al., 2000).

Sequencing

The sequencing step relies on the NGS technologies. The three most popular, massively parallel NGS platforms, are currently dominating the NGS market and widely used in RNA-seq, including the 454 pyrosequencing system (a subsidiary of Roche), the AB SOLiD system (Life Technologies), and the Illummina Genome Analyzer (Illumina) (Liu et al., 2012a; Marguerat and Bähler, 2010). All these three NGS platforms rely on an in vitro cloning step (clonal amplification) to amplify each fragmented cDNA molecule in a cell-free system, because their sensitivities are not high enough for the single molecule sequencing (Metzker, 2009). Specifically, both the 454 and the SOLiD systems employ an innovative emulsion PCR. In the emulsion PCR, the cDNA fragments from a library are attached to beads and subsequently compartmentalized in the aqueous droplets of a water-in-oil emulsion such that each droplet contains a single DNA molecule; the segregated template fragments are then amplified in the tiny aqueous droplets of the emulsion (Dressman et al., 2003). Different from the 454 and the SOLiD systems, the Illummina Genome Analyzer performs a so-called bridge PCR amplification, in which the adapter-linked, single-stranded cDNA fragments are first immobilized on a glass slide by oligonucleotide hybridization in a bridging way, followed by clonal PCR amplification (Adessi et al., 2000; Fedurco et al., 2006). Clonal amplification results in a population of identical templates, each of which is subjected to the following sequencing reaction. Due to PCR artifacts, clonal amplication may introduce bias in the RNA-seq results as well. One way to discriminate PCR artifacts is to perform different biological replicates and determine whether same short reads are concurrently present in different replicates (Wang et al., 2009).

NGS platforms use different sequencing strategies (Metzker, 2009). The sequencing mechanism employed by Roche 454 is pyrosequencing, which is a nonelectrophoretic, bioluminescence-based method (Metzker, 2009; Ronaghi et al., 1998). In brief, every bead, containing the clonally amplified template originating from a single cDNA molecule, is transferred to a ∼29 μm well on a Picotiter Plate (a fiberoptic chip). A mixture of enzymes, such as DNA polymerase, ATP sulfurylase, and luciferase, are also packed into each well. The four DNA nucleotides are added sequentially in a fixed order across the Picotiter Plate during a sequencing run. When one or more complementary nucleotides are added, it generates a light signal that is recorded by the camera in the instrument. The signal strength is proportional to the number of nucleotide added. For the SOLiD system, a mechanism, known as sequencing by ligation (SBL), is used (Landegren et al., 1988; Metzker, 2009; Shendure and Ji, 2008). In its simplest form, after clonal amplification, a sequencing primer is hybridized to template with 3′ at position ‘n’. A set of four fluorescently labeled di-base-encoded probes (8 bp oligonucleotides) then compete for ligation to the sequencing primer. The best-matching probe is linked to the primer by DNA ligase, followed by removal of nonligated probes by wash. Fluorescence imaging is then done to determine the identity of the ligated probe. Following a series of ligation cycles, the extension product is removed and the template is reset with the second sequencing primer with 3′ at the ‘n-1′ position for a second round of ligation cycles. After five rounds of primer reset, each base is examined twice by two different primers, although SOLiD is hence not running as fast as other methods. Last, the method of cyclic reversible termination (CRT) is used by IIIumina (Bentley et al., 2008). CRT is a cyclic sequencing-by-synthesis method that differs from SBL in its use of nucleotide monomer and DNA polymerase. In particular, four types of reversible terminator bases are added in the presence of DNA polymerase and template, and after incubation non-incorporated nucleotides are washed away. Imaging is then performed to determine the identity of the incorporated nucleotide, and, then the dye and the terminal 3′ blocker are chemically removed, allowing for the next cycle to begin. In addition to CRT, ion semiconductor sequencing (Life Technologies) uses a method of “sequencing by synthesis” as well, which is based on the detection of hydrogen ions released during the polymerization of DNA. The major benefits of this method are rapid sequencing speed and low operating costs (Rothberg et al., 2011). Several reviews describing the mechanisms and comparing the advantages and disadvantages of these NGS technologies were published elsewhere (Ansorge, 2009; Liu et al., 2012a; Metzker, 2009; Shendure and Ji, 2008).

Despite the popularization of the NGS technologies, the so-called third generation sequencing methods, also known as single-molecule sequencing methods, are on their way. These methods, such as Heliscope sequencing and single-molecule real-time (SMRT) (Pacific Biosciences), are featured by omission of template clonal amplification and real-time signal capture (Liu et al., 2012a). Heliscope sequencing has already been used for a published RNA-seq study (Ozsolak et al., 2010). Additionally, the single-molecule direct RNA sequencing (DRS) technology, developed by Heliscope, is emerging. DRS sequences RNA molecules directly in a massively-parallel manner without biasing sample manipulations such as RNA reverse transcription, ligation, and clonal amplification (Ozsolak and Milos, 2011a). However, these technologies are at varying stages of development and suffer from many drawbacks, such as low single-pass accuracy (81%∼83%), low sequencing efficiency, and low throughput (Niedringhaus et al., 2011; Schadt et al., 2010).

Bioinformatic analysis

After the signal processing, NGS platforms generate millions of short sequences, termed reads, associated with their base-call quality scores that indicate the reliability of each base call. The lengths of these short reads are within a range of 25–450 bp, depending on the type of NGS platform. The resulting reads are categorized into three types: exonic reads, exon–intron junction reads, and poly(A) reads (Fig. 2) (Wang et al., 2009). Although the application of NGS technologies in RNA-seq makes transcriptomic sequences handy, the analysis of the huge amount of RNA-seq data are still the bottleneck for understanding the transcriptome. Fortunately, during the past few years, various bioinformatic tools (software) for RNA-seq data analysis have been developed. Especially, the significance of Bioconductor should be emphasized here. Bioconductor is a free, open source, and open development software project, mainly based on the R programming language, for the analysis and annotation of RNA-seq data and other types of high-throughput genomic data (Gentleman et al., 2004). Most Bioconductor components are organized as R packages. These Bioconductor packages and other software for RNA-seq data analysis have been well reviewed elsewhere (Chen et al., 2011; Febrer et al., 2011; Oshlack et al., 2010). Nowadays, researchers can routinely combine these tools to form the best analysis pipeline per their research interests.

FIG. 2.

Short reads mapping and quantification of the digital signal (modified from Wang et al., 2009). The resulting reads are aligned against a reference genome and fall into three groups, including exonic reads (green), junction reads (red-brown), and poly(A) end reads (blue) (Wang et al., 2009). These reads are used to generate an expression profile with a single-base resolution.

A typical analysis pipeline of RNA-seq data is outline in Figure 1. Quality assessment is the first step for the RNA-seq data analysis. To ensure a coherent final result, low-quality sequences, over-represented sequences, and adapter sequences have to be filtered out, and this step can be accomplished with Bioconductor packages, such as ShortRead (Morgan et al., 2009) and Biostrings (Pages et al., 2009). Once high-quality reads have been obtained, these short reads are subsequently aligned to a reference genome or transcriptome. Alignment is usually done with software outside Bioconductor. In contrast to conventional alignment algorithms, these software are based on the indexing strategies that are able to align millions of short reads in a reasonable period of time (Langmead et al., 2009). Basically, these aligners fall into two categories: one is based on the Burrows-Wheeler transform algorithm, such as Bowtie (Langmead et al., 2009) and BWA (Li and Durbin, 2010), and the other based on Needleman-Wunsch or Smith-Waterman algorithm, such as GNUMAP (Clement et al., 2010), BFAST (Homer et al., 2009), and SHRiMP (Rumble et al., 2009). The aligners in the first category is much faster, the aligners in the second category, however, despite more time needed, are usually more sensitive and generate more reads correctly aligned (Garber et al., 2011). Since many reads span exon–exon junctions and therefore cannot be directly aligned, specialized aligners (spliced aligners), such as Erange (Trapnell et al., 2009) and IsoformEx (Kim et al., 2011), are developed to split the junction reads and then independently align the split read fragments. The methods available to study RNA splicing from short RNA-Seq data have been well reviewed elsewhere (Alamancos et al., 2013). When the reference genome or transcriptome is unavailable, de novo transcriptome assembly can be performed (Robertson et al., 2010). De novo assembly is used for most fish studies using RNA-seq because only limited fish species have the whole genome information. De novo assembly software, for instance, ABySS (Simpson et al., 2009), Velvet (Zerbino and Birney, 2008), and Trinity (Grabherr et al., 2011), use a de Bruijn graph approach, which aligns the user-defined sequence overlap (referred as k-mer) between two reads to create contigs (Grabherr et al., 2011). Transcriptome de novo assembly is impeded by the repeats within huge amounts of short reads, alternatively spliced transcripts, as well as biased transcriptome sequencing coverage (Liu et al., 2012b). Therefore, different improvements for the de Bruijn graph approach, such as using various k-mer lengths instead of a single one, have been made to optimize transcriptome assembly (Surget-Groba and Montoya-Burgos, 2010). Besides, a reference proteome or genome from an evolutionarily linked species can be used to aid de novo assembly.

Once all reads have been appropriately filtered and mapped or assembled, they can then be counted, and gene expression levels can thus be inferred from the total counts of reads belonging to the exons of a particular gene. In this way, an expression score can be assigned to every base, leading to a genome-scale transcriptome map in terms of quality and quantity (Fig. 2). The single-base (digital) resolution of RNA-seq allows for detection of gene expression at the isoformic and allelic levels and discovery of previously unannotated genes (Jiang and Wong, 2009; Levin et al., 2009; Oshlack et al., 2010; Trapnell et al., 2010). RNA-seq analysis allows not only quantifying gene expression levels within a single RNA sample but also detecting differential expression (DE) across treatments or conditions (Kvam et al., 2012; Oshlack et al., 2010); the latter is usually the real interest of most studies. However, for the DE analysis of RNA-seq data, normalization has to be performed to adjust for between-sample differences such as library size and within-sample gene-specific features regarding GC content and gene length (Dillies et al., 2012; Kvam et al., 2012). There is no standard method to detect DE due to the short history of RNA-seq technology. The currently popular tools for DE analysis in Bioconductor include edger (Robinson et al., 2010), DESeq (Anders and Huber, 2010), baySeq (Hardcastle and Kelly, 2010), and tweeDEseq (Esnaola et al., 2013), and methods, not included in Bioconductor, also exist, such as ShrinkBayes (Van De Wiel et al., 2013) and TSPM (Auer and Doerge, 2011). These DE analysis tools have already been reviewed in detail and well compared in their gene ranking performances (Dillies et al., 2012; Kvam et al., 2012; Soneson et al., 2013). In addition, RNA-Seq analysis can also enable us to detect SNPs, fusion genes, and post-transcriptional gene regulation, such as RNA editing, RNA degradation, and RNA translation (Marguerat and Bähler, 2010).

New Biological Insights Gained in Fish Studies Using RNA-seq

The advent of RNA-seq technology has begun to revolutionize fish transcriptomic studies. In the past several years, RNA-seq has been applied to a number of studies involved with various fish species, including not only model species, such as zebrafish (Collins et al., 2012), but also commercially important and eco-environmentally relevant fish species, such as channel catfish (Ictalurus punctatus) (Liu et al., 2012b), European sea bass (Dicentrarchus labrax) (Sarropoulou et al., 2012), and rainbow trout (Oncorhynchus mykiss) (Palstra et al., 2013). Although the application of RNA-seq in fish transcriptomics is at the nascent stage, the results of a Pubmed literature search (with keywords, “fish” and “RNA-seq”) indicated that the number of publications in this field has increased considerably in the last 3 years. And, to our knowledge, no comprehensive review paper has appeared on this subject. Therefore, the aim of this section is to review the literature on fish transcriptomic studies using RNA-seq.

Transcriptome mapping and genome annotation

The high-throughput RNA-seq technology, independent of prior knowledge, allows efficient transcriptome annotation, including, for instance, transcript start and end sites and the identification of novel transcripts (Wang et al., 2009). For example, using RNA-seq, Liu and colleagues sequenced and de novo assembled the doubled haploid channel catfish transcriptome and generated 370,798 non-redundant transcript-derived contigs (Liu et al., 2012b). Functional annotation of these contigs revealed 25,144 unique protein-encoding transcripts. Of these 225,144 unique transcripts, over 14,000 transcripts were identified as full-length transcripts with complete open reading frame, and about 90% of these full-length transcripts were identified with the complete 5′ and 3′ ends. The lengths of 5′ UTRs (∼254 bp) were found to be much shorter than those of 3′ UTRs (∼1,096 bp). Similarly, reference transcriptomes have been constructed for other fish species, including zebrafish (Collins et al., 2012), guppy (Poecilia reticulata) (Fraser et al., 2011), and turquoise killifish (Nothobranchius furzeri) (Petzold et al., 2013)’

Creating a comprehensive reference transcriptome using RNA-seq provides invaluable information for genome annotation, such as the gene and exon boundaries, as well as the identification of novel transcribed regions. For example, to improve zebrafish genome annotation, RNA-seq data, based on an optimized analysis pipeline, was used to adjust intron/exon boundaries of the defined genes, confirm their expression, and improve the coverage of 3′ untranslated regions of genes (Collins et al., 2012). This optimized pipeline can potentially be applied to improve genome annotation for other organisms.

Novel transcript discovery

RNA-seq has been used to discover novel transcribed regions in the genome. The results from RNA-seq suggest the existence of a large number of unknown transcribed regions in every fish species surveyed, including zebrafish (Pauli et al., 2012), rohu carp (Labeo rohita, Hamilton) (Robinson et al., 2012), and rainbow trout (Palstra et al., 2013). Among these unknown transcripts, many nc RNAs other than rRNA and tRNA are of great interest due to their essential roles in many cellular processes, including translation, RNA splicing, gene regulation, and genome defense (Mattick and Makunin, 2006). These ncRNAs can be broadly classified as either long ncRNAs (lncRNAs; >200 nucleotides) or small RNAs (sRNAs; <200 nucleotides). More than 550 distinct long intervening noncoding RNAs (lincRNAs, one type of lncRNAs) were identified in zebrafish embryo (Ulitsky et al., 2011); 224 unique mature microRNAs (miRNAs, one type of sRNAs) were identified in Atlantic salmon (Salmo salar) (Kure et al., 2013). These newly discovered transcripts will facilitate the annotation of sequenced genome and lay the ground for future functional studies.

Interrogation of post-transcriptional modification (RNA splicing)

The sequence of the mature mRNA molecule can differ substantially from the corresponding genome sequence. This is because the precursor mRNA undergoes three main post-transcriptional modifications, including 5′ capping, 3′ polyadenylation, and RNA splicing, in the cell nucleus before the translation process begins. RNA splicing has been studied in fish using RNA-seq (Aanes et al., 2011; Pauli et al., 2012). Analysis of zebrafish transcriptome dynamics during maternal to zygotic transition by RNA-seq revealed the frequency of alternative splicing in zebrafish embryo to be in a range between 50% and 60%, which is considerably high, but much lower when compared to the splicing frequency estimated for human, ranging from 92% to 95% (Aanes et al., 2011). Another RNA-seq study in zebrafish embryo identified 3532 transcripts as the variants of known RefSeq genes (novel isoforms and partial transcripts), supplementing the existing exon–intron structures of many genes in zebrafish (Pauli et al., 2012). These extensive splicing variants, in conjunction with many novel transcribed regions, give strong evidence of previously unappreciated transcriptome complexity.

In addition to identification of splicing isoforms, RNA-seq was used to study the process of alternative splicing regulation in zebrafish. A genome-wide and target-specific role of U1C protein in 5′ splice-site recognition and selection was characterized, adding this protein to the growing list of splicing regulators (Rösel et al., 2011).

SNP discovery

SNP is a single nucleotide variation at a given position in the genome between members of a biological species. SNPs are widely distributed throughout the genome and have been extensively used as the markers for many applications in genomics and genetics. With its superior sensitivity and single-base resolution, RNA-seq has been proved to be a very effective tool for the identification of gene-associated SNPs at a genome-wide scale. Several SNP identification-directed RNA-seq studies in fishes have already been reported. To developing SNP arrays in catfish, Liu and colleagues conducted RNA-seq in multiple individuals of both channel catfish and blue catfish (Ictalurus furcatus) (Liu et al., 2011). With the help of the SNP calling module in CLC Genomics Workbench (CLC bio, Aarhus, Denmark), as well as a SNP quality screening procedure, they identified 342,104 intra-specific SNPs for channel catfish, 366,269 intra-specific SNPs for blue catfish, and 420,727 inter-specific SNPs between channel catfish and blue catfish; these SNPs were found to be distributed within 16,562 unique genes in channel catfish and 17,423 unique genes in blue catfish. In another study, RNA-seq was performed to discover gene-associated SNPs in four strains of common carp (Cyprinus carpio) (Xu et al., 2012). BWA and SAMtools software were applied to align the reads to reference transcriptome and call SNPs, and, in total, 712,042 intra-strain SNPs and 53,893 inter-SNPs were identified. These identified SNPs provide a solid base for the future genetic studies in these fish species and will contribute to the development of a high throughput SNP genotyping platform.

Since 90% of the genetic difference between individuals is explained by SNPs, SNPs serve as invaluable markers for selection of important traits in breeding (Collins et al., 1998). In one RNA-seq study regarding identification of SNP markers for growth traits, 22 SNP markers and one mitochondrial haplotype were found to be significantly associated with growth traits in rainbow trout (Salem et al., 2012). In addition, SNPs are widely used as markers for distinguishing allelic transcripts when studying allele-specific expression that is essential for normal development and many cellular processes (Bell and Beck, 2009). RNA-seq provides a perfect tool for allele-specific expression studies by identifying SNPs and quantifying transcripts at the same time. For example, RNA-seq was successfully performed to assess the allele-specific expression in a F1 interspecies hybridized from southern platyfish (Xiphophorus maculates) and monterrey platyfish (Xiphophorus couchianus) (Shen et al., 2012).

Quantifying transcript level

One superior advantage of RNA-seq is its quantitative nature, enabling researchers to capture the transcriptome dynamic changes in response to environment or to intrinsic programs. Almost every RNA-seq studies published so far has included this analysis, and they all suggest that RNA-seq data gives reliable measurements of transcript levels within or between samples. In the last few years, RNA-seq, with a striking speed, has been widely used to detect differential gene expression in fish studies relating to developmental biology, immunology, evolutionary biology, physiology, toxicology, and diseases.

Developmental biology

Zebrafish is used as a model system, with the most prominent application in developmental biology. So far, the application of RNA-seq in the studies of fish developmental biology has been confined to zebrafish. RNA-Seq was employed to compare the transcriptome profiles of four early developmental stages (1-, 16-, 512-cell stage, and 50% epiboly) in zebrafish on a global scale (Vesterlund et al., 2011). In this study, only 177 genes were detected as developmentally regulated, while a majority of gene transcripts were present at a steady level, and a major transition in gene regulation and transcriptional activity took place between the 512-cell and 50% epiboly stages. To determine the role of vitamin D receptor (VDR) in zebrafish embryogenesis, transcriptome dynamics were assessed using RNA-seq in zebrafish embryos/larvae treated with 1α,25(OH)₂D₃ (active metabolite of vitamin D₃) or vehicle for various periods of time (Craig et al., 2012). The expression levels of genes for transcription factors, peptide hormones, receptor-activator of NFκB ligand (RANKL) and of genes encoding proteins that plays key roles in fatty acid, amino acid, and xenobiotic metabolism pathways were significantly affected, demonstrating that 1α,25(OH)₂D₃ regulates multiple pathways in zebrafish embryogenesis. RNA-seq was also used to study zebrafish retinogenesis, and Id2a protein was identified as an intrinsic regulator of retinogenesis that balances between proliferation and differentiation during retinogenesis by modulating Notch pathway gene expression (Uribe et al., 2012).

Immunology

Comparison of the fish transcriptomes before and after immune challenges leads to the identification and subsequent characterization of immune-related genes and specific pathways involved in the immune responses, helping to create immune-based therapy for fish diseases, select disease-resistant fish brood stocks, and understand the origin and evolution of immune system. RNA-seq has been used in this kind of study for live fish, fish embryo, and fish primary cells.

The RNA-seq studies regarding fish immune responses to pathogens are mainly performed in economically important species. For instance, the transcriptomic response of channel catfish gill to Flavobacterium columnare, a Gram-negative bacterium implicated in fish disease outbreaks worldwide, was investigated using Illumina sequencing (Sun et al., 2012). Using a 1.5-fold change cut-off, 2605 uniquely annotated genes, with critical roles in pathogen recognition, cytoskeletal dynamics, cell junction integrity, oxidative stress responses, apoptosis, lysosomal processes, and pro- and anti-inflammatory pathways, were found to exhibit significant differential expression patterns. Fifteen differentially expressed genes detected by RNA-seq were confirmed by quantitative PCR (qPCR). In this study, a rhamnose-binding lectin (RBL) gene was highlighted, with a 105-fold increase in expression level. This discovery led to a subsequent study from the same group, in which RBL ligands, L-rhamnose and D-galactose, strongly protected channel catfish against columnaris disease in a dose-dependent manner (Beck et al., 2012). The same group also investigated the transcriptomic change in the intestinal epithelium of channel catfish following Edwardsiella ictaluri challenge (Li et al., 2012). In this study, 1633 differentially expressed genes, implicated in actin cytoskeletal polymerization/remodeling and junctional regulation in pathogen entry and subsequent inflammatory responses, were identified. Similar studies were carried out in Asian seabass (Lates calcarifer) (Xia et al., 2013) and Japanese sea bass (Lateolabrax japonicus) (Xiang et al., 2010), both of which were challenged with Vibrio harveyi.

To clarify the host immune mechanisms underlying the protective effects of vaccines and improve its immunogenicity in the future efforts, RNA-seq was used to investigate the immunization-related gene expression patterns of zebrafish and European sea bass immunized with vaccines against Edwardsiella tarda and Vibrio anguillarum, respectively. In the study with zebrafish, 4565 genes were expressed differentially in liver transcriptome samples before and after immunization (2186 up-egulated and 2379 down-egulated) (Yang et al., 2012). Further qPCR analysis confirmed that the genes encoding the factors involved in major histocompatibility complex (MHC)-I processing pathway were upregulated, while those involved in the MHC-II pathway were downregulated. In the RNA-seq study with European sea bass, differential expression was detected for 496 transcripts in head kidney and for 336 in gut (Sarropoulou et al., 2012).

Aside from in vivo studies, the fish immune response was also investigated in fish embryo and primary cells using RNA-seq. The innate host immune response to inflammatory bacterial infection was probed with zebrafish embryos infected with Salmonella typhimurium using both RNA-seq and tag-based sequencing (Ordas et al., 2011). In this study, two sequencing methods showed a strong correlation of sequence read counts per transcript and an overlap of 241 transcripts differentially expressed in response to infection. These transcripts were found to encode transcription factors, signal transduction proteins, cytokines and chemokines, complement factors, proteins involved in apoptosis and proteolysis, proteins with anti-microbial activities, as well as many known or novel proteins not previously linked to the immune response. Additionally, RNA-Seq analysis of Poly (I:C) (polyinosinic:polycytidylic acid)-challenged rainbow trout erythrocytes revealed diverse groups of differentially expressed mRNA transcripts related to multiple physiological systems including the endocrine, reproductive, and immune systems (Morera et al., 2011).

Aquatic toxicology

Aquatic toxicology aims to elucidate the effects of toxic chemicals on aquatic organisms at multiple levels, from subcellular level through individual organisms to communities and even ecosystems. Fish is one of the most significant research subjects in aquatic toxicology. RNA-seq enables researchers, from the transcriptomic perspective, to deduce how organisms respond to environmental pollutants. On the basis of such fundamental knowledge, the ultimate goal in aquatic toxicology is to predict and thus diminish or prevent the harmful effects of aquatic pollutants on the environment.

Oleksiak and colleagues (2011) utilized RNA-seq, in conjunction with a previous microarray data, to probe the transcriptomic response to polychlorinated biphenyl (PCB) exposure in embryos and larvae from a PCB-sensitive population and a PCB-resistant population of Atlantic killifish (Fundulus heteroclitus). Differential expression analysis revealed a sizeable set of PCB-responsive genes in the sensitive population, a much smaller set of PCB-responsive genes in the resistant fish, and few similarities in PCB-responsive genes between the two populations. The RNA-seq data corroborated most of the microarray results and detected novel transcripts that were not captured by microarray. This discrepancy in results between microarray and RNA-seq was also observed in other researches and might be improved by statistical methods with proper assessment the statistical significance of the observed changes (Esnaola et al., 2013). In another study, the toxicological effects of perfluorooctane sulfonate (PFOS), a widely-distributed persistent organic pollutant, on Oryzias melastigma embryos were examined using RNA-seq (Huang et al., 2012). Of the 145,394 genes detected by RNA-seq, 325 genes were significantly upregulated, while 349 genes were significantly downregulated, and these differentially expressed genes were found to be implicated in neurobehavioral defects, mitochondrial dysfunction and the metabolisms of proteins and fats. In a most recent study, RNA-seq was used to identify miRNAs in Atlantic salmon muscle tissue as potential biomarkers of toxicological stress (Kure et al., 2013). A total of 18 miRNAs were significantly differentially expressed in response to acidic aluminum-rich water, 4 downregulated and 14 upregulated. These identified differential expressed miRNAs may have the potential as biomarkers for other fish species as well. Other similar studies using RNA-seq were carried out in brown trout (Salmo trutta) (Uren Webster et al., 2013), Gulf killifish (Fundulus grandis) (Garcia et al., 2012), and yellow perch (Perca flavescens) (Pierron et al., 2011) stressed by a mixture copper and zinc, deepwater horizon oil, and a mixture cadmium and copper, respectively. In general, these studies have successfully shown that RNA-seq is a powerful approach to study the ecotoxicological response of fish to the polluted environments.

Physiological processes

Stress triggers certain physiological responses and plays an essential role in the natural selection. Due to the volatile aquatic environment in the wild, fish are exposed to different kinds of stressors, including lack of food, extreme water salinity, extreme temperature, pathogens, toxicants, and many others. Physiological responses to stresses are manifested by the changes in gene expression. So, identifying differentially expressed genes and pathways under stress conditions is important for the understanding of stress responses in fish. Stress responses to pathogens and toxicants have been discussed in the foregoing sections, and we thus focus on other stressors here.

The transcriptomic response of the liver of crimson-spotted rainbowfish (Melanotaenia duboulayi) to elevated temperature (21°C versus 33°C) was studied using RNA-seq (Smith et al., 2013). In this study, of the 107,749 assembled transcripts, 4251 transcripts were differentially expressed, and over 1000 of these differentially expressed transcripts were annotated; in addition to the well-characterized temperature-responsive genes relating to protection against apoptosis or maintenance of protein structure, gene ontology analysis revealed many novel temperature-responsive genes involved in catabolism, lipid metabolism, and oxidoreductase activity, indicating increased metabolism to cope with the increased temperature and the resulting hypoxic conditions. Analogously, RNA-seq was applied to determine the heat stress-induced gene expression profile in channel catfish (25°C versus 36°C), revealing similar results as crimsonspotted rainbowfish (Liu et al., 2013). These studies suggest that fish possess great plasticity in dealing with environmental temperature variations and assist the development of heat-tolerant fish strains for aquaculture. In addition, RNA-seq was performed to study the effects of the temperature variations during embryonic development on the adult thermal adaption ability in zebrafish (Scott and Johnston, 2012). Aside from heat stress, transcriptomic responses to salinity and fasting have been studied in Asian seabass (Xia et al., 2013).

Furthermore, RNA-seq studies regarding other aspects of fish physiology, such as swimming (Palstra et al., 2013) and circadian clock (Tovin et al., 2012), have been reported as well.

Evolution biology

Elucidation of the molecular basis of population divergence and speciation is one of the major challenges in evolutionary biology. Gene expression divergence is thought to be one of the mechanisms underlying phenotypic divergence. During the past few years, RNA-seq has become an invaluable tool for the study of phenotypic divergence by allowing whole transcriptome sequencing. For instance, RNA-seq analysis was conducted to study the adaptive transcriptomic divergence between dwarf and normal lake whitefish (Coregonus clupeaformis spp., Salmonidae) (Jeukens et al., 2010). In this study, normal whitefish was found to overexpress the genes related to protein synthesis, while dwarf fish overexpress the genes associated with immunity, DNA replication and repair, and energy metabolism, and the correlation between RNA-seq results and a previous microarray data was positive. Transcriptomic analyses using RNA-seq were similarly employed to study the genetic bases for the phenotypic differentiation between siscowet and lean lake trout (Salvelinus namaycush) (Goetz et al., 2010), for the divergent pigment patterns in marine and freshwater sticklebacks (Gasterosteus aculeatus) (Greenwood et al., 2012), for the adaptation of Mexican tetra (Astyanax mexicanus) to the cave environment (Gross et al., 2013), and for the homologous relationship between zebrafish swimbladder and mammalian lung (Zheng et al., 2011). In these transcriptomic studies, many differentially expressed genes were detected, laying the ground for future functional genomic studies investigating the heritable genetic changes governing the phenotypic divergence in the evolutionary process.

Conclusion

RNA-seq technology is still in its infancy stage, however, it has superior advantages over other transcriptomic approaches, such as microarray, tilling array, and tag-based sequencing approaches. In the past few years, RNA-seq has made substantial contributions to our understanding of the fish transcriptome in terms of transcriptomes annotation, determination of the transcriptional structure of genes, and quantitative analysis of transcriptome dynamics during different biological processes. As RNA-seq technology continuously evolve and its cost keeps decreasing, within the next few years RNA-seq will without doubt be exploited to a larger extent and lead to many more exciting discoveries regarding the transciptomics of fish and other organisms.

Abbreviations Used

CRT

cyclic reversible termination

DE

differential expression

DRS

direct RNA sequencing

lincRNA

long intervening non-coding RNA

lncRNA

long non-coding RNA

miRNA

microRNA

MPSS

massively parallel signature sequencing

mRNA

messenger RNA

ncRNA

non-coding RNA

NGS

next-generation sequencing

PMAGE

polony multiplex analysis of gene expression

RBL

rhamnose-binding lectin

RNA-seq

RNA sequencing

rRNA

ribosomal RNA

SAGE

serial analysis of gene expression

SBL

sequencing by ligation

SMRT

single molecule real time

SNP

single nucleotide polymorphism

sRNA

small RNA

tRNA

transfer RNA

WTSS

whole transcriptome shotgun sequencing

Acknowledgments

Preparation of this review was supported by Shanghai University Knowledge Service Platform Project (ZF1206).

Author Disclosure Statement

The authors declare that no conflicts of interest exist.